1

The determinants of pricing in the Mexican domestic airline sector: the impact of competition and airport congestion

Agustin J. Ros1

NERA Economic Consulting

200 Clarendon St., 11th fl.

Boston, MA 02116-5089

617-927-4506

617-927-4501 fax

Agustin.ros@nera.com

1 Vice President NERA Economic Consulting, agustin.ros@nera.com. This work evolved from the OECD-Mexico Competition Assessment Toolkit project between the OECD and the Mexican Competition Commission while the author was a Senior Economist, Competition Division of the OECD. The opinions expressed and arguments employed herein do not necessarily reflect the official views of the OECD or of the governments of its member countries, nor the Mexican Competition Commission.

2

Abstract: Low-cost airlines in Mexico affect the lowest-quoted fares of one of the two principal incumbent carriers, but have no effect on the lowest-quoted fares of the other incumbent carrier. The same conclusion holds for competition between incumbent carriers where the lowest-quoted fares of one of the incumbent carriers is lower when incumbents compete. Congestion at the Mexico City airport is limiting potential competition, with carriers being able to charge a significant price premium. This suggests that the societal costs of airport congestion can go well beyond the negative congestion externality and should also include the effects of reduced competition. These findings raise important public policy issues.

Classification JEL: L40, L93, R41.

Key Words: Airlines, Airport, Airport Congestion, Competition, Low-cost Carriers, Mexico.

I. Introduction Low-cost carriers entered the Mexican domestic airline market in 2005, and by 2008 they

provided service to approximately one-third of domestic passengers (Secretariat of

Communications and Transport (SCT), 2008). The presence of low-cost airlines on a route has

been found to be associated with lower airline fares and significant consumer benefits. Morrison

(2001) found that in 1998 the low-cost carrier Southwest Airlines was responsible for $12.9

billion in consumer savings — approximately 20 percent of the industry’s 1998 domestic

scheduled passenger revenue. Bennet and Craun (1993) found that average fares were

approximately 50 percent lower in markets served by Southwest Airlines. Daraban and Fournier

(2008) found that incumbents dropped fares by a total of 22 percent prior to and after entry by

Southwest, a finding that is consistent with Goolsbee and Syverson (2008).

In 2005 the SCT declared the Mexico City airport to be officially “saturated,” which

means that the number of airlines using takeoff and landing slots is at a maximum during certain

times of the day and no further flights can takeoff or land. Thus, some airlines cannot enter the

3

Mexico City airport to compete against the established carriers that already operate at that airport,

and the expected impact of this should be higher fares. When an airport becomes saturated, the

government is empowered to implement reforms to alleviate the levels of congestion by, for

example, auctioning the slots. Congestion at airports has been found to impose significant costs

on consumers. Daniel and Harback (2009) found that implementing congestion pricing at 27

major US airports would reduce delays by 13 passenger years and 1000 aircraft-hours each day,

saving $3-5 million daily. Other studies have also examined the impact of airport congestion on

welfare; see, e.g., Morrison (1983) and Morrison and Winston (1989). These studies, however,

have not examined the impact that airport congestion can have on competition and on airline

fares. Abramowitz and Brown (1993) examine price determinants in the US airline industry

controlling for the effects of congestion, as measured by the takeoffs and landing per runway at

the two endpoints, and find that congestion is associated with higher fares.

Using a new dataset that is based on advertised, lowest-fare quotes, this paper extends the

literature by examining the impact of low-cost carriers on domestic lowest-quoted airfares in

Mexico and the impact that saturation conditions in the Mexico City airport are having on lowest-

quoted fares.2 The paper also examines the impact that competition between the two incumbent

carriers has on their lowest-quoted fares.

Section II provides a summary of the liberalization process in Mexico and a review of

market structure. Section III describes the data that were used for this study and provides some

2 The use of lowest-quoted fares in this study presents limitations in interpretation of the results, as any finding that incumbent carriers’ lowest-quoted fares are affected by the presence of low-cost carriers does not necessarily mean that incumbent carriers’ average fares would exhibit the same impact. In the extreme, it may be the case that the incumbents have dropped their fares for only one round trip, or offer a very limited number of seats with the lowest-quoted fare, in order to be able to advertise "with fares as low as..." but kept the rest of their fares high, in which case the welfare effects of competition may be modest. Unfortunately, the data do not permit one to link lowest-quoted and average fares in a market.

4

summary statistics. Section IV presents econometric models (both OLS and instrumental variables

estimations), and Section V presents conclusions and policy recommendations.

II. Liberalization and Market Structure The two principal airlines in Mexico are Aeromexico and Mexicana, which are referred to

throughout this paper as the “incumbent” carriers because they have been operating for over 70

years — much longer than any other current airline in Mexico. Although these airlines started out

as private enterprises (Mexicana in 1921 and Aeromexico in 1934), by the early 1980s the

government owned majority shares in both airlines. The companies were privatized in the late

1980s when they served approximately 90 percent of the market. Entry into the sector was

regulated by the government through the issuing of concessions and permits to provide air service.

That process, however, was neither transparent nor predictable, and obtaining a concession to

enter and compete against the two incumbent carriers was not easy. In August 1991 the

government eased its restrictions on entry, which led to the creation of new airlines that competed

against Aeromexico and Mexicana.3

By 1995 Aeromexico and Mexicana were in difficult financial conditions due to increased

competition and the financial crisis that affected Mexico in that year. These circumstances forced

the government, once again, to take majority control of both companies, a situation that would last

more than ten years until the “second” privatizations of Mexicana in 2005 and Aeromexico in

2007.

In the mid 2000s — the period coinciding with the privatization of Mexicana and

Aeromexico — the industry experienced an important development: the entry of low-cost carriers.

3 See Mexican Federal Competition Commission “Caso Cintra” available at http://www.cfc.gob.mx.

5

Carriers such as Interjet, Vivaaerobus, and Volaris entered the market in the period 2005-2006

and since then have established themselves as important competitors. Other carriers following the

low-cost carrier model — such as Alma and Avolar — also entered the market but have since

exited.

A competition analysis of the domestic airline sector in Mexico — based upon publicly-

available data (SCT 2008) — leads to several conclusions about the sector’s performance and

level of competition: The sector experienced fairly robust growth during the period for which

data are available, 1989-2008. The compound annual growth rate for total domestic passengers

between 1989 and 2008 was approximately 5.4 percent; this was a period during which Mexico’s

real annual GDP growth was approximately 3.0 percent.

There has been a significant decrease in concentration during the same period with strong

recent gains made by the three low-cost carriers. In 1989 the Herfindalh-Hirschman Index

(“HHI”) — based upon a nationwide market and output measured by number of passengers —

was 4396; by 2008 the nationwide HHI had decreased to 1766.4 Five low-cost carriers entered the

market in 2005-2006, and by the end of 2008 the three low-cost carriers that remained in the

market had captured more than 30 percent of passenger traffic.

There is little publicly-available information on profitability. The evidence that exists

indicates low levels of profitability for the two incumbent operators since the mid 2000s.5 And,

since 2007 four airlines have exited the market.

With respect to pricing, the government does not publish time-series data on domestic

airline prices in Mexico. Based upon two analyses of 12 domestic routes conducted by a Mexican 4 Author’s calculation based upon data in SCT (2008). 5 See World Airline Reports 2005-2008, available at http://www.atwonline.com.

6

think tank — one in 2000 and one in 2008 — airline prices decreased by approximately 4 percent

per year between those two time periods; see aregional.com (2008). Based upon the study of

lowest-quoted fares in this paper, however, prices on those same routes were close to 20 percent

higher in 2009 compared to 2008.6

III. Data Description & Summary Statistics The SCT provides passenger information by route — monthly and yearly passengers travelled and

total flights — for 589 point-to-point routes in Mexico in 2008.7 These 589 routes roughly equal

the population of routes in Mexico. For this study, data were obtained for 497 of these routes.8

Price data for the airlines were collected during the period April through August 2009.9

In 2008 the SCT data show a total of 13 airlines (both regional and trunk) providing

domestic service in Mexico.10 Of the 13 airlines, three ceased operating sometime in 2008 (these

airlines were Aerocalifornia, Alma, and Avolar). This left ten airlines for data analysis. In 2008

the SCT data shows only four regional airlines; this is a number that is significantly lower than in

earlier years. Of the four airlines, however, two are owned by the two large incumbent trunk

airlines, Aeromexico and Mexicana. Aeromexico operates Aeroliteral (Aeromexico Connect), 6 Obtaining airline prices requires a standard methodology in order to compare results across time. In this study, the lowest price, round-trip fares with all taxes and surcharges included were selected. For some routes, the departure date was less than a week, while for other routes it was more than two weeks with a Saturday night stayover. The aregional.com report does not detail the methodology that they used to collect prices for these routes. Caution is therefore urged in comparing the results and reaching conclusions. 7 See SCT (2008). 8 The difference between the 589 routes identified in 2008 and the 497 routes for which data were collected can be explained, in part, by seasonality — some of the 589 routes are not flown during the months that the data were collected — and by the possibility that some of the routes are no longer being serviced. A review of the data show that these missing routes were proportionally lower demand routes than the average route for which data were obtained for this study. The mean for these missing routes was 159 flights per year compared to the mean of 887 flights per year for this study. 9 This time period is very atypical for the Mexican economy and the airline sector in Mexico. Mexico was hit very hard by the U.S. recession in 2008-2009. And the H1N1 virus epidemic was first reported in Mexico in late April 2009 and subsequently caused a severe contraction in travel to, from and within Mexico. Most of the data (approximately 95 percent) were collected in July and August of 2009. 10 Trunk carriers tend to fly larger aircraft and provide service throughout Mexico, while regional carriers tend to fly smaller aircraft and provide air travel within discrete geographical areas of Mexico.

7

and Mexicana operates Aerovias Caribe (Click). These two regional carriers provided service to

more than 85 percent of the passengers flying on regional routes as determined by the SCT. The

data analysis below does not distinguish between the incumbent operators — Aeromexico and

Mexicana — and their respective regional subsidiaries — Connect and Click, respectively.11

Thus eight airlines were left for data analysis when data collection commenced.12

For each route identified by the SCT the web sites of each of the airlines were consulted to

determine whether the airline provided service on that route.13 If the airline offered service for

that route, a fare quote was obtained from the airline’s website for the lowest quoted round-trip

service.14 For each airline that provided service on the route an attempt was made to obtain fare

quotes for the same departing/returning dates for all carriers. When this was not the case, the

difference in date was generally within plus or minus one calendar day. This occurred in less than

10 percent of the observations. Thus, for each route there is a lowest fare quote for each airline

providing service on the route. These individual carrier lowest-quoted fares as well as an

unweighted lowest available average fare across all carriers serving the route were used as the

dependent variables in the various models estimated below.

Some additional comments about the lowest-quoted fare variables are in order. First, the

departure and return date were used to create dummy variables for each route (and thus each

11 Aeromexico and Mexicana did not have separate web sites for their respective subsidiaries, and it was not always clear for a given route whether the flight would be provided by the subsidiary. Therefore, fares in this study for Aeromexico and Mexicana represent fares for flights offered by these airlines including flights offered by their respective subsidiaries. 12 One airline, Aviacsa, ceased providing passenger service shortly after data collection began. The airline was shut down by the government for a number of reasons; although in press statements the airline expects to provide service again, as of the end of our data collection it had not resumed service. 13 All of the airlines had a functioning website that allowed customers to search different itineraries and make reservations. 14 Fares included all applicable taxes. Where a carrier offered only one-way flights the data were not collected — this was the case for some Interjet and Volaris observations. The routes include some that some carriers fly non-stop and some that are connecting flights.

8

observation) that indicated whether lowest-quoted fares were discriminating between business and

leisure travelers. To create this variable there are three relevant dates: the data collection date, the

departure date, and the return date. If the departure and return date were taken 2-3 weeks from

the data collection date and required a Saturday night stay, the fare quote was considered as one

for a leisure passenger. Since lowest fare quotes were collected from each airline for the same

departure and return date, each observation has a dummy variable (“leisure”) indicating whether

the departure date for the route was greater than two weeks and required a Saturday night stay.15

Another dummy variable (“non-leisure”) measured whether the departure date was less than 7

days from the data collection date and did not require a Saturday night sleep-over.16 The choice in

the departure and return date, and thus whether the route is considered a leisure or non-leisure

route, was random.

Second, when an option between a non-stop and one-stop flight was given on an airline

site the former option was selected as long as the fares were the same. If they were not the same,

then the lowest fare option was selected.17

Third, the route A-B-A is considered a different route than B-A-B. For example, one

observation was for the route Mexico City to Cancun to Mexico City and another observation was

15 Unfortunately data were not collected for whether the departure and return dates were 2-3 weeks in advance and required a Saturday night stay for the first 60 route observations. We treat these as missing observations in the regression models below, which explains why the leisure variable has 437 observations while many of the other variables have 497 observations. Thus, the 60 observations for these routes are not included in the structural models estimated in Tables 5 and 6 below. 16 Data for this variable were collected for all the 497 routes in the sample. 17 Flights with two stops were not considered even though they might have presented a cheaper alternative (this was seldom the case with Mexicana). The reason for this was the unlikely preference of a customer to travel within Mexico with two stops (some flights like this lasted several hours) and the substantial reduction in the average fare that this choice entailed.

9

for the route Cancun to Mexico City to Cancun. We found that fares on these routes were

somewhat different.18

Table 1 presents the variable names and data sources that are used in the study as well as a

description of each variable.

TABLE 1 VARIABLES USED IN STUDY

Variable Name Variable Description Source Fare 1 Lowest-quoted fare by each airline on each

route, in Mx pesos per kilometer Respective airlines’

websites Fare 2 Average lowest-quoted fare per route for all

airlines offering service on the route, in Mx pesos per kilometer

Respective airlines’ websites

Distance Distance in kilometres of non-stop roundtrip travel between origin and destination

www.world-airpot-codes.com

pass08 Number of passengers in 2008 for the month the data were collected (in 000)

SCT

gdp07 2007 nominal GDP per capita of the origin city in Mx pesos (in 000)

INEGI & CONAPO

Nonleisure Dummy variable, 1 if data collected less than 1 week before departure,

N/A

Leisure Dummy variable, 1 if data collected more than 2 weeks in advance and requires a

Saturday night stay

N/A

Touristdest Dummy variable, 1 if destination is a tourist destination

Consejo de Promocion Turistica,

http://www.cptm.com.mx/index.jsp, and own research

Mexico City O Dummy variable, 1 if origin airport is Mexico City, only airport that is “saturated”

N/A

Mexico City D Dummy variable, 1 if destination airport is Mexico City, only airport that is “saturated”

N/A

Airportcost The average of the two (origin & destination) airport charges in pesos in 2007 covering the cost of using the airport, takeoff/landing fees,

platform and security costs

Comision Federal de Competencia

Ncomp Number of airlines offering service on the Own research 18 This could be due to airlines’ charging different fares for these routes — i.e., charging different fares for roundtrips from Mexico to Acapulco and Acapulco to Mexico. It could also be due to the fact that airline fares for this study were collected over a 4-5 month period.

10

route Nlcc Number of “low-cost” carriers offering

service on the route Own research

Lcc Dummy variable, 1 if a low-cost carrier offers service on the route

Own research

incumbentcomp Dummy variable, 1 if there is competition between the two incumbent airlines,

Aeromexico and Mexicana on the route

N/A

N/A: not applicable

Table 2 presents the descriptive statistics for the variables used in the study. Some of the

more interesting findings are those on mean fares, mean number of competitors (Ncomp) per

route, mean number of low-cost competitors (Nlcc) per route, the presence of at least one low-cost

competitor (Lcc) per route, whether the two main incumbent carriers compete (Incumbentcomp),

and whether the destination or origin airport is Mexico City.

TABLE 2 DESCRIPTIVE STATISTICS

Variable Name Obs Mean Std. Dev. Min Max Fare 1 1094 2.81 2.20 0.51 15.41 Fare 2 497 2.87 2.09 0.64 14.29 Distance 497 916 611 58 3,233 distance (lcc)(1) 207 1125 635 214 3233 distance (non-lcc)(2) 290 768 549 58 3171 pass08 497 4.84 10.55 0 90.29 gdp07 497 120.96 86.80 41.94 776.943 Nonleisure 497 0.47 0.50 0 1 leisure(3) 437 0.54 0.50 0 1 Touristdest 497 0.25 0.43 0 1 Mexico City O 497 0.10 0.30 0 1 Mexico City D 497 0.10 0.30 0 1 Airportcost 491 1666.92 132.41 1375 1939 Ncomp 497 2.20 1.14 1 7 Nlcc 497 0.55 0.76 0 3 Lcc 497 0.42 0.49 0 1 Incumbentcomp 497 0.48 0.50 0 1

(1) Average distance flown by low-cost carriers; (2) Average distance flown by non-low-cost carriers; (3) There are only 437 observations for the leisure variable because we did not track this for the first 60 observations collected.

11

The data show that mean lowest quoted airline fares are approximately $2.80 Mx

peso/kilometer19 and that the mean number of competitors on the average route in Mexico is 2.2.20

The mean value of the number of low-cost competitors on a route is 0.55, and a low-cost

competitor was present on 42 percent of the routes analyzed. Competition between the two

incumbent carriers, Aeromexico and Mexicana, occurred on 48 percent of the routes, and 10

percent of the routes involved the Mexico City airport as the origin airport and 10 percent as the

destination airport; this is the airport that was saturated as defined by the SCT. Approximately 25

percent of the flights were to tourist destinations.

Table 3 below presents summary lowest-quoted fare statistics by airline. The three low-

cost airlines are Interjet, Vivaaerobus, and Volaris while the “traditional” carriers are

Aeromexico, Mexicana, Aeromar, Aviacsa, and Magnicharters.21 Low-cost carriers’ lowest-

quoted fares per kilometer are significantly lower than the lowest-quoted fares per kilometer of

the traditional carriers, although as the table reveals low-cost carriers tend to fly longer routes.

Each low-cost carrier’s mean lowest-quoted fare is well below the lowest-quoted fares of the

traditional carriers. The average lowest-quoted fare of the incumbent carriers is 3.05 pesos per

kilometer compared to 1.78 pesos per kilometer for low-cost carriers, a statistically significant

19 Applying an exchange rate of 13 Mx pesos per US$ and converting kilometers to miles, one obtains a figure of approximately $US 0.35/mile. According to Morrison and Winston (2008), in 2007 U.S. revenue per passenger mile was approximately $US 0.13/mile. Thus, average airline prices in Mexico in 2009 were almost three times higher than average airline prices in the U.S. in 2007. Of course demand and cost characteristics, including average route distances and the level of competition, in the U.S. and in Mexico vary greatly and can help explain the difference. 20 Morrison and Winston (2008) found the average number of effective competitors per route in the U.S. in 2007 to be approximately 2.3. 21 The term traditional is meant to distinguish these airlines from the low-cost carrier model. Interjet, Vivaaerobus, and Volaris have followed the low-cost carrier model which tends to have a different cost structure, a common fleet, and operates out of secondary airports. See Gillan (2006) for a discussion of low-cost carrier model.

12

difference of approximately 42%. The lowest-quoted fares of the two largest traditional airlines,

Aeromexico and Mexicana, are almost identical. The airline with the highest lowest-quoted fare

is Aeromar, a traditional carrier, while the airline with the lowest-quoted fares is Vivaerobus, a

low-cost carrier.

TABLE 3 DESCRIPTIVE LOWEST-QUOTED FARE STATISTICS BY AIRLINE AND

TYPE OF CARRIER

Obs Average distance/route

Mean Fares/km

Std. Dev.

Min Max

“Traditional” Carriers Aeromexico 354 964 2.88 1.90 0.75 15.41

Mexicana 347 866 2.98 1.37 0.95 15.24 Aeromar 97 677 5.15 3.09 1.25 14.36 Aviacsar 10 869 2.90 1.37 0.95 5.20

Magnicharter 13 803 2.45 0.97 1.46 4.40 Average 3.05

“Low-cost” Carriers Interjet 145 977 2.01 0.91 0.90 4.70

Vivaaerobus 46 934 1.32 0.45 0.71 2.72 Volaris 82 1496 1.39 0.44 0.70 2.99 Average 1.78*

*Statistically significant difference at the 1% level.

V. Models

In this section several models are estimated to investigate the impact on lowest-quoted fares from

low-cost carriers, competition between incumbents, and airport saturation conditions in Mexico

City. The first model is non-structural in the sense that all observable and non-observable route-

specific characteristics are captured through route-specific dummy variables, with the exception

13

of incumbent competition and low-cost carrier impact on incumbent lowest-quoted fares and some

interaction effects that are explicitly accounted for in separate dummy variables.22 The model is:

(1) yir = β0 + βiXi + βrXr + βlccXlcc + βIncXInc + βlccintXlccint + βIncintXIncint + εi ,

where yir is the natural log of carrier i’s lowest quoted fare on route r23; β0 is a constant term; Xi

and βi are a vector of carrier dummy variables and coefficients, respectively (i=8); Xr and βr are a

vector of route dummy variables and coefficients, respectively (r=497); Xlcc and XInc are two

dummy variables if the observation belongs to a route where there is a low-cost carrier (Xlcc) or

where there is competition between the two incumbents (XInc), respectively; Xlccint and XIncint are

four interaction dummy variables formed by multiplying the Aeromexico and Mexicana dummy

variable by the low-cost carrier dummy and by the incumbent competition dummy variable; and εi

is the residual.24 Model 1 is estimated via ordinary least squares. Table 4 below contains the

results showing only the coefficients of the 8 airlines, the coefficients for the low-cost carrier and

incumbent competition dummy variables, and the interaction coefficients. The coefficients of the

approximately 500 route dummy variables are not reported in Table 4.

TABLE 4 REGRESSION OF LOG OF INDIVIDUAL LOWEST-QUOTED FARES ON CARRIER &

ROUTE-SPECIFIC DUMMY VARIABLES

Coefficient Aeromexico .036

(.149) Mexicana -.25*

(.131) Aeromar -.142

(.118) 22 The other models to be estimated are structural models that explicitly account for observable route-specific information. 23 Fare 1 as described in Table 1 is used as the dependent variable. 24 In different specifications we also included dummy variables for the month that the outbound flight occurred and did not find any material difference in the results.

14

Aviacsa -.468**

(.163) Magnicharters -.563***

(.122) Interjet -.533***

(.109) Vivaaerobus -.885***

(.119) Volaris -.480***

(.116) Lcc -.053

(.181) Incumbentcomp -.189

(.170) Aeromexico*LCC dummy -.032

(.087) Mexicana*LCC dummy -.188*

(.086) Aeromexico*Incumbent comp dummy -.359***

(.079) Mexicana*Incumbent comp dummy -.129

(.071) Constant .978***

(.209) R-sq .894

N 1094 Numbers in parenthesis are standard errors; *p<.05, **p<.01, ***p<.001

The interaction effects of the low-cost carrier dummy variable and the two airline dummy

variables indicate that Mexicana is more affected by the presence of low-cost carriers than is

Aeromexico. Mexicana’s lowest-quoted fares are approximately 17 percent lower on its routes

that have a low-cost carrier, a finding that is statistically significant at the 5 percent level. By

contrast Aeromexico’s lowest-quoted fares are only 3 percent lower on its routes that have a low-

cost carrier, but the finding is not statistically significant. With respect to the interaction effects of

the incumbent competition dummy and the two airline dummies, Aeromexico’s lowest-quoted

fares are approximately 30 percent lower on routes where it competes with Mexicana.

15

Interestingly, the same does not apply for Mexicana. Mexicana’s lowest-quoted fares are not

significantly different on routes where it competes with Aeromexico. Both Lcc and

Incumbentcomp are negative but not significant, indicating that average lowest-quoted fares on

routes with a low-cost carrier or with incumbent competition are not different compared to when

there is not a low-cost carrier or incumbent competition.

Finally, the results in the table can be used to calculate the relative lowest-quoted fares of

carriers, holding constant all route-specific characteristics, thus making the comparisons more

meaningful than a simple comparison of averages. Interpreting the magnitude of the coefficient is

not straightforward given the approximately 500 route-specific dummy variables. What is

relevant, however, is a comparison of each carrier’s coefficient vis-à-vis the other carriers.

According to the results, Aeromexico has the highest lowest-quoted fares while Vivaaerobus has

the lowest-quoted fares. If we hold constant all route-specific characteristics, the low-cost carrier

average impact on lowest-quoted fares is approximately -46 percent, while the incumbent impact

on lowest-quoted fares is approximately -23 percent, or a difference of approximately 25 percent.

Thus, whereas the data in Table 3 showed that low-cost carrier lowest-quoted fares were

approximately 40 percent lower, when one controls for all route-specific characteristics the figure

drops to approximately 25 percent.

Structural models were also estimated to examine the main determinants of domestic

lowest-quoted fares, controlling for observable, route-specific characteristics and including

several variables that are important to policy makers. The following general model was

estimated:

(2) Y = Xβ + u ,

16

where Y, the dependent variable, is the lowest-quoted fare per route of the individual carriers or,

depending on the specific model, the average lowest-quoted fare per route of the carriers offering

service on the route, X is an N x k matrix of sample values of the independent variables, β are the

k parameters to be estimated, and u is the stochastic disturbance with the assumption that E [u | x]

= 0; i.e., the unobserved factors in the regression function are not related systematically to the

observed factors.25

The main policy variables of interest are: (1) the presence of low-cost carriers on a given

route (Lcc) – when the dependent variable is the non-LCC carriers’ lowest-quoted fares (i.e.,

Aeromexico, Mexicana or Aeromar) or the average lowest-quoted fares; (2) the impact that

saturation conditions in the Mexico City airport have on lowest-quoted fares captured through two

Mexico City dummy variables; and (3) the impact of competition (Incumbentcomp) between the

two incumbent airlines Aeromexico and Mexicana. The sign and magnitude of these variables

can provide evidence on the potential gains that can be expected from implementing sound

competition policy in the sector. Policymakers can influence the presence of low-cost carriers on

routes, perhaps indirectly, by enforcing competition law and enacting sound competition policies

that remove impediments to the entry and expansion of competitive airline suppliers. And the

effects of competition between the two incumbent airlines provide valuable information in the

event that the airlines were to attempt to merge.

With respect to the impact of saturation conditions in the Mexico City airport, Mexico

City is different in many ways from other city airports, and some of these factors are controlled

for in our structural models through other independent variables, such as average distance per

route, income, population, etc. In addition, the variable (airportcost) is used to control for the 25 Further below this assumption is relaxed, and parameters are estimated based upon the instrumental-variables estimator.

17

maximum price that airports can charge for services such as takeoff/landing fees, platform and

security costs, etc. Adding this variable controls for the fact that airline fares are higher when

airports charge higher prices, ceteris paribus and airport charges should be higher on routes where

opportunity costs are higher — i.e., in those airports that are congested.

Nevertheless, even though the models control for these observable route-specific

information, it is still possible that the Mexico City dummy variable is picking up additional

effects that are not necessarily related to saturation conditions, i.e., because of Mexico City’s

“special status”. Thus, it would not be correct to attribute the entire Mexico City price premium

to airport saturation conditions. In order to further control for this, the Mexico City dummy

variable is split into two dummy variables: one if the roundtrip route has the origination in Mexico

City (Mexico City O), and the other if the roundtrip route has the destination in Mexico City

(Mexico City D). The demand for roundtrip travel into Mexico City would more likely pick up

this Mexico City “special status” effect, while roundtrip travel out of Mexico City is likely free of

such effects. Thus, a finding of similar coefficients for Mexico City O and Mexico City D

supports the hypothesis that Mexico City premium is congestion related. If so, this suggests that

the Mexico City dummy variables are picking up: (1) the impact of very high barriers to entry at

the Mexico City airport, and (2) the lack of potential competition at the Mexico City airport —

two competition policy concerns that policymakers can address by implementing market-based

mechanisms (e.g., auctioning off slots, implementing peak-load slot pricing, etc.) to alleviate slot

congestion at the Mexico City airport and ensure that slots are utilized and assigned in the most

efficient manner.

18

Other control variables in model (2) include dummy variables for airlines Aeromar and

Magnicharters,26 the inverse of distance in kilometers between the origin and destination city (on

the assumption that price is likely not linearly related to distance), the income of the origin city

(2007 GDP per capita measured in thousands of Mexican pesos), the geometric mean of the

endpoint populations, a leisure travel dummy variable, and a tourist destination dummy variable.27

Table 5 presents the results.

TABLE 5 MODELS OF AVERAGE & AIRLINE LOWEST_QUOTED FARES IN MEXICO, OLS

ESTIMATES Dependent Variablesa

Aeromexico Fares

Mexicana Fares

Aeromar Fares

Interjet Fares

Average Faresb

1/Distance 241.07***

(57.99) 115.61***

(37.97) 276.43***

(54.24) 300.04***

(53.45) 148.15***

(25.24) Gdpcap07 .000034

(.00024) -.00046 (.00031)

-.00034 (.00032)

.00008 (.00017)

-.00029 (.00023)

Pop -.00012*

(.00006) -.00020***

(.00007) -.00014*

(.00007) -.00012*

(.000052) -.00016***

(.000048) Leisure -.0089

(.0469) .0126

(.0596) -.2110*

(.0951) -.0727 (.0516)

.0104 (.0402)

Touristdest -.0996 (.0596)

-.0641 (.0712)

.0191 (.1805)

.1329

(.0718) -.0587 (.0514)

Mexico City O .3988***

(.1060) .7702***

(.1588) .4653*

(.2115) .3187**

(.1223) .5374***

(.1041) Mexico City D .3031***

(.0963) .6015***

(.1587) .5092**

(.2146) .2969**

(.1210) .4399***

(.0997) Airportcost .00068***

(.00024) .00068

(.00035) .00017

(.00052) .00039

(.00031) .00054**

(.00023) Incumbentcomp -.1495*** -.0657 -.0943 .0518 -.0403

26 Different model specifications were attempted that included the number of competitors (Ncomp) on a route as an independent variable and included the LCC and Incumbentcomp variables. The results of Ncomp were counter-intuitive showing a positive relationship between fares and Ncomp, possibly due to the fact that the model contained separate and overlapping measures of competition. Model (2) now contains four different sources of competition that do not overlap: one from the low-cost airlines (Lcc), one from the incumbent airlines (Incumbentcomp), and one each from airlines that are significantly different: Aeromar and Magnicharters. As stated in footnote 12, Aviacsar ceased operating shortly after the date collection began. 27 In different specifications we also included dummy variables for the month that the outbound flight occurred. The only material difference was that the Mexico City O variable was no longer statistically significant at the 5% level for the model where Aeromar is the dependent variable.

19

(.0506) (.0781) (.0982) (.0558) (.0378) Lcc -.0417

(.0476) -.2338***

(.0594) -.0452 (.1274)

-.2098***

(.0384) Aeromardummy .1541*

(.0766) .4431***

(.0707) .1280

(.0792) .4411***

(.0588) Magnidummy -.0303

(.0854) -.2840*

(.1251) -.0835 (.1336)

-.1107 (.1079)

-.2210***

(.0924) _cons -.4531

(.4155) -.4041 (.6025)

.7595 (.8692)

-.4906 (.5674)

-.1921 (.3842)

F-stat 7.49*** 15.87*** 12.68*** 13.28*** 26.93***

R-sq .488 .458 .604 .543 .505 Rmse .386 .484 .399 .285 .406

N 311 292 81 125 431 Numbers in parenthesis are standard errors; *p<.05, **p<.01, ***p<.001. a All fares are in Mexican pesos/km. b The dependent variable used is Fare 2; see Table 1.

Five models were estimated.28 The first four models have as the dependent variable the

individual carrier’s lowest-quoted fare while the fifth model has as the dependent variable the

average lowest-quoted fare on the route for carriers offering service on the route.

Heteroskedasticity was detected in each of the four models; thus the coefficients’ standard errors

were estimated using the Huber-White-sandwich estimator of the variance.

The strongest findings in Table 5 are those related to the Mexico City dummy variables:

the saturation conditions in the Mexico City airport. The coefficients of the Mexico City O and

Mexico City D dummy variables are similar, positive, and strongly significant, thus consistent

with the hypothesis that the premiums that are found for roundtrip flying into or out of Mexico

City in the models are related to saturation conditions at the Mexico City airport. Because the

model controls for observable route-specific characteristics, including the cost of using the airport,

the Mexico City premium is likely due to the existence of very high entry barriers and very weak

28 The aeromar dummy variable is dropped when aeromar is the dependent variable as there is no variation in the variable since it is always a one; similarly, the low-cost carrier dummy variable is dropped when Interjet is the dependent variable.

20

potential competition. Relying on the coefficient of Mexico City O as the best measure of the

impact of airport saturation (for the reason discussed above that the fares for roundtrip flights

originating in Mexico City are likely free of the special status effect), the results indicate that

there is a high price premium that is associated with the saturation condition that greatly limits the

ability of potential competition to discipline fares. The impact also varies by airline, with

Mexicana being able to charge the highest premium in Mexico City, approximately 116 percent

and Aeromexico charging approximately a 50 percent premium. Saturation conditions at the

Mexico City airport are also affecting the lowest-quoted fares of Interjet, the only low-cost carrier

that operates at the airport. Interjet, is able to charge approximately a 38 percent premium

because of the lack of potential competition in Mexico City. The average impact, based on the

model with average fares as the dependent variable, is approximately 71 percent.

With respect to the impact that low-cost carries have on carriers’ lowest-quoted fares,

Table 5 indicates that the impact is greatest on Mexicana but practically non-existent for

Aeromexico or Aeromar; this is the same finding that is found in Table 4 from estimating model

(1). Mexicana’s fares are approximately 20 percent lower and statistically significant on those

routes where a low-cost carrier is present. Aeromexico’s and Aeromar’s lowest-quoted fares are

approximately 4 percent lower on those routes where a low-cost carrier is present, but they are not

statistically significant. Thus, the evidence on the impact that low-cost carriers’ lowest-quoted

fares are having on the lowest-quoted fares of the incumbent operator is mixed. Low-cost

carriers’ lowest-quoted fares are affecting the lowest-quoted fares of Mexicana, thus providing

evidence that the low-cost carriers and Mexicana are in the same relevant product market. The

evidence for including Aeromexico and Aeromar and the low-cost carriers in the same product

market, however, is not as strong. As the results from Table 4 indicated, low-cost carriers’

21

lowest-quoted fares are approximately 25 percent below the average incumbent lowest-quoted

fares, controlling for all route-specific characteristics. The evidence indicates, however, that only

Mexicana is responding to the lower-quoted fares that are offered by the low-cost carriers.

Competition between incumbent airlines (i.e., between Aeromexico and Mexicana) has an

impact on the lowest-quoted fares of Aeromexico but no impact on the lowest-quoted fares of

Mexicana. According to Table 5, Aeromexico’s lowest-quoted fares are approximately 14 percent

lower on those routes where it competes with Mexicana, somewhat less than the 30 percent found

in Table 4 from model (1). Mexicana’s lowest-quoted fares are approximately 6 percent lower on

such routes, but the result is not statistically significant.

Additional strong findings in the model are related to distance and population, both of

which reduce fares per kilometer. In addition, lowest-quoted fares are higher on Interjet tourist

destinations (by approximately 17 percent) and lower on Aeromar’s fares that are considered for

leisure travelers (by approximately 19 percent).29

Using OLS to estimate the parameter values in equation (2) results in unbiased parameter

estimates if E [u | x] = 0; i.e., if the unobserved factors in the regression function are not related

systematically to the observed factors. The presence of a low-cost carrier on a route may be

endogenous in equation (2).30 For example, it is possible that the decision of carriers to enter a

particular route is based, in part, on the average prices that exist for the route. All else equal,

29 Because of the “uniqueness” of Mexico City, the models in Table 5 were also all estimated dropping the Mexico City observations in order to assess how sensitive the results were to the Mexico City observations. Dropping the Mexico City observations did not affect the major conclusions found in Table 5 with respect to the impact of Lcc or incumbent competition; the results were actually stronger without the Mexico City observations. 30 Incumbentcomp is a strategic decision of the airlines that may be based in part on unobserved factors that affect the profitability of these routes and may also be correlated with the error term. Thus, treating this variable as exogenous might result in biased parameter estimates.

22

profits on a route are positively correlated with higher prices so we would expect to observe a

positive relationship between the existence of carriers on a route and prices.31 Unobservable

shocks in equation (2) reflected in u would affect not only Y but Lcc as well and thus result in

biased parameter estimates. In order to determine the robustness of the conclusions described

above, instrumental variables (IV) are used to estimate equation (2).32

In order to use IV to estimate equation (2), instruments for the potentially endogenous

explanatory variable (Lcc) are required. To derive consistent estimates of (2) the instruments

must be highly correlated with Lcc and yet be uncorrelated with u. Two instruments are used for

Lcc. The first is the share of routes at the two endpoint airports that are served by low-cost

carriers. That is, the routes that are part of the two endpoint airports are summed, and the percent

of those routes that have at least one low-cost carrier is used as the instrument. A regression of

Lcc on this variable was positive and highly significant. The second instrument that is used is the

number of carriers that serve the endpoints. That is, for each endpoint of a route the number of

carriers that serve that airport is determined, and the two numbers are added. A regression of Lcc

on this variable was positive and highly significant. Table 6 presents the results of the IV

estimation.33

31 Interestingly, using logit to regress LCC on the natural log of prices results in a negative and significant relationship, holding constant distance, gdpcap, and the number of passengers. 32 It is possible, however, that lcc in equation (2) is only weakly endogenous and the use of IV may not result in better estimates. The decision by carriers to enter a particular route will likely be influenced, in part, by average prices for that route. And, all else equal, higher average prices should lead to greater profits and provide a greater incentive to enter the route. Over time, however, we would expect rents to be competed away, and in equilibrium the relationship between prices and low-cost carriers may be weak. Unfortunately, we do not have data on how long a low-cost carrier has been providing service on the route in question. In addition, as discussed in the previous footnote, the relationship between price and lcc was negative. Moreover, the decision of a carrier to enter a particular route will more likely be influenced by the previous period prices than by current prices. 33 In different specifications we also included dummy variables for the month that the outbound flight occurred. The only material difference was that the Mexico City O variable was no longer statistically significant at the 5% level for the model where Aeromar is the dependent variable. In addition, the model with Interjet as the dependent variable is excluded from Table 6 because there is no variation in the potentially endogenous variable LCC when Interjet is the dependent variable.

23

TABLE 6 MODELS OF AVERAGE & AIRLINE LOWEST_QUOTED FARES IN MEXICO, IV

ESTIMATES Dependent Variablea

Aeromexico Fares

Mexicana Fares

Aeromar Fares

Average Faresb

1/Distance 260.40***

(59.66) 126.12***

(35.05) 284.84***

(44.36) 138.29***

(26.16) gdpcap07 .00014

(.00024) -.00023 (.00032)

-.00020 (.00029)

-.00018 (.00024)

Pop -.00012*

(.00006) -.00017**

(.00007) -.00011 (.00007)

-.00012***

(.000046) Leisure -.0299

(.0474) .0227

(.0605) -.2083**

(.0881) .0220

(.0405) Touristdest -.0864

(.0588) -.0382 (.0690)

.1006 (.1751)

-.0602 (.0510)

Mexico City O .3837***

(.1027) .7278***

(.1638) .4291*

(.1981) .5076***

(.1037) Mexico City D .2830***

(.0949) .5830***

(.1611) .5088**

(.1998) .4137***

(.0994) Airportcost .00060**

(.00025) .00075*

(.00035) .00025

(.00048) .00063***

(.00024) Incumbentcomp -.1659***

(.0496) -.0253 (.0748)

-.0422 (.0891)

-.0240 (.0387)

Lcc -.0322 (.0855)

-.4636***

(.1322) -.1812 (.1666)

-.3680***

(.0800) Aeromardummy .1867**

(.0735) .3977***

(.0669) .4114***

(.0573) Magnidummy -.0336

(.0831) -.2524

(.1433) -.1437 (.1316

-.2013

(.1049) _cons -.3305

(.4164) -.5855 (.5894)

.5728 (.8139)

-.3048 (.3849)

F-stat 8.08 17.71 12.28 27.70 Rmse .381 .485 .374 .406

N 311 292 81 431 Hansen J statistic Chi-sq(1) P-val

9.923 0.0016

2.437 0.1185

1.02 0.3125

0.614 0.4332

Numbers in parenthesis are standard errors; *p<.05, **p<.01, ***p<.001. a All fares are in Mexican pesos/km. b The dependent variable used is Fare 2; see Table 1.

24

The estimates from the models in Table 6 are efficient for arbitrary heteroskedasticity and

include robust standard errors.34 The Hansen J statistic does not reject the hypothesis that the

instruments are orthogonal to μi (and thus uncorrelated with the error term) except for the

Aeromexico model. Hausman specification tests, however, do not reject the null hypothesis that

OLS estimators are consistent and efficient.

None of the conclusions reached from the OLS models are changed as a result of using IV

estimates, although the point estimates of some of the coefficients change slightly. This provides

evidence, as confirmed through the Hausman specification tests, that the OLS estimators are

consistent and efficient. The biggest change in magnitude is the impact that low-cost carries have

on Mexicana’s lowest-quoted fares. The OLS model showed approximately 20 percent lower

quotes, while the IV estimate is approximately 37 percent lower. The impact of airport saturation

is similar in both models. Whereas the average impact in the OLS model was 71 percent, the

average impact in the IV model is 66 percent. And the impact that incumbent competition is

having on Aeromexico’s lowest-quoted fares is similar in both models — approximately 14

percent in the OLS model and approximately 15 percent in the IV model.

V. Conclusions & Policy Recommendations The entrance of low-cost carriers into the domestic airline sector in Mexico in 2005 has

significantly benefitted consumers, as exhibited by the growth of passengers flying and declines in

industry concentration since 2005. When we control for route-specific characteristics, low-cost

carriers’ lowest-quoted fares are approximately 25 percent lower than the incumbent carriers’

34 The generalized method of moments (GMM) estimator was used for producing robust standard errors. The results do not change in any material way if the presence of i.i.d. errors is assumed.

25

lowest-quoted fares. And, the presence of low-cost carriers on a route is associated with

reductions in the lowest-quoted fares of one of the two principal incumbent carriers.

Given the fact that low-cost carriers have captured approximately one-third of the market,

the welfare benefits to Mexican consumers from low-cost carrier entry are likely significant.

Policymakers can and should remove unnecessary restrictions to carriers entering the domestic

market. This can be done by eliminating the SCT’s discretion in awarding concessions and

approving new routes so that any safe airline can get a concession. Increasing the limit of foreign

ownership and negotiating agreements with other countries to permit cabotage service would also

increase competition and benefit Mexican consumers.

By far, the most important reform that policymakers can implement is to deal with the

saturation conditions in the Mexico City airport, which is acting as a barrier to entry and removing

the ability of potential competition to discipline fares. This study shows that average lowest-

quoted fares are approximately 70 percent higher because of such conditions. Both incumbent

airlines already operating in Mexico City as well as the only low-cost carrier operating in Mexico

City are able to charge a premium due to the absence of potential competition. Policymakers

should implement market-based solutions — e.g., auctions for takeoff/landing slots, etc. — to

allocate takeoff and landing slots and modify the pricing rules at the Mexico City Airport so that

they are transparent and improve economic efficiency by, for example, taking into account

external congestion costs and/or other factors within a price-cap plan to limit monopoly pricing.

More can be done to create a level playing field for entrants at the airport, such as considering

eliminating “grandfather” clauses in the current regulations that favor the incumbent operators’

access to essential airport infrastructure and modifying the regulations so that actual and potential

entrants are represented in the committees that administer and allocate the takeoff/landing slots.

26

Acknowledgments The author would like to thank members of the Technical Committee of the OECD-Mexico

Competition Assessment Toolkit project for their many insights and suggestions. I also thank Douglas Umana for his

excellent assistance in data gathering and analysis and Craig Conrath, Sean Ennis, and Ernesto Estrada for their many

insights and suggestions. I especially thank Severin Borenstein, Gary Dorman, Jose Gomez Ibanez, Steve Morrison,

Bernie Reddy, William E. Taylor, Timothy Tardiff, and Lawrence J. White for their review and significant

contributions to the analysis as well as two anonymous referees. All errors are my responsibility.

References Abramowitz, Amy D. and Brown, Stephen M. (1993). Market share and price determination in the

contemporary airline industry. Review of Industrial Organization, 8, 419-433. Aregional.com. (2008). Competencia en el mercado del transporte aéreo de pasajeros en México.

in Competencia en sectores estratégicos de México available from http://www.aregional.com/mexico/?target=estatal&lang=en&PHPSESSID=PHPSESSID=#.

Bennett, R. and Craun, J. M. (1993). The airline deregulation evolution continues: the southwest effect. Office of Aviation Analysis, U.S. Department of Transportation. Retrieved January 7, 2011 from www.ostpxweb.dot.gov/aviation/X...files/Southwest%20Effect.DOC.

Daniel, J. and Harback, K. T. (2009). Pricing the major U.S. hub airports. Journal of Urban Economics, 66, 33-56.

Daraban, B. and Fournier, G. M. (2008). Incumbent responses to low-cost airline entry and exit: a spatial autoregressive panel data analysis. Research in Transportation Economics, 24, 15-24.

Gillan, D. (2006). Airline business models and networks: regulation, competition and evolution in aviation markets. Review of Network Economics, 4, 366-385.

Goolsbee, A. and Syverson, C. (2008). How do incumbents respond to the threat of entry? evidence from major airlines. Quarterly Journal of Economics, 123, 1611-1633.

Morrison, S. A. (1983). Estimation of long-run prices and investment levels for airport runways. Journal of Transportation Economics, 1, 103-130.

Morrison, S. A. (2001). Actual, adjacent, and potential competition: estimating the full effect of southwest Airlines. Journal of Transport Economics and Policy, 35, 239-256.

Morrison, S. A. and Winston, C. (1989). Enhancing the performance of the deregulated air transportation system. Brookings Papers on Economic Activity, Microeconomics, 61-123.

Morrison, S.A. and Winston, C. (2008). The state of airline competition and prospective mergers. Hearings Before the Judiciary Committee Antitrust Taskforce, United States House of Representatives, April 24, 2008. Retrieved January 7, 2011 from http://www.brookings.edu/~/media/Files/rc/testimonies/2008/0424_airlines_winston/0424_airlines_winston.pdf.

27

Secretariat de Comunicaciones y Transportes (2008), La Aviacion Mexicana en Cifras, 1989-2008. Retrieved January 7, 2011 from http://www.sct.gob.mx/transporte-y-medicina-preventiva/aeronautica-civil/estadisticas/.